---
license: apache-2.0
base_model: openai/whisper-small
tags:
- generated_from_trainer
datasets:
- common_voice_6_1
model-index:
- name: whisper-small-ta-fyp
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# whisper-small-ta-fyp

This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the common_voice_6_1 dataset.
It achieves the following results on the evaluation set:
- eval_loss: 0.5417
- eval_wer: 65.8901
- eval_runtime: 1043.495
- eval_samples_per_second: 1.707
- eval_steps_per_second: 0.214
- epoch: 10.55
- step: 2500

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
- mixed_precision_training: Native AMP

### Framework versions

- Transformers 4.39.0.dev0
- Pytorch 2.1.0+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2